sklearn-docs/MNIST_classification_using_multinomial_logistic_L1

1import plotly.express as px2import numpy as np3 4from sklearn.datasets import fetch_openml5from sklearn.linear_model import LogisticRegression6from sklearn.model_selection import train_test_split7from sklearn.preprocessing import StandardScaler8from sklearn.utils import check_random_state9import gradio as gr10 11 12# Load data from https://www.openml.org/d/55413X, y = fetch_openml(14    "mnist_784", version=1, return_X_y=True, as_frame=False, parser="pandas"15)16 17print("Data loaded")18random_state = check_random_state(0)19permutation = random_state.permutation(X.shape[0])20X = X[permutation]21y = y[permutation]22X = X.reshape((X.shape[0], -1))23 24 25scaler = StandardScaler()26 27 28def dataset_display(digit, count_per_digit, binary_image):29    if digit not in range(10):30        # return a figure displaying an error message31        return px.imshow(32            np.zeros((28, 28)),33            labels=dict(x="Pixel columns", y="Pixel rows"),34            title=f"Digit {digit} is not in the data",35        )36 37    binary_value = True if binary_image == 1 else False38    digit_idxs = np.where(y == str(digit))[0]39    random_idxs = np.random.choice(digit_idxs, size=count_per_digit, replace=False)40 41    fig = px.imshow(42        np.array([X[i].reshape(28, 28) for i in random_idxs]),43        labels=dict(x="Pixel columns", y="Pixel rows"),44        title=f"Examples of Digit {digit} in Data",45        facet_col=0,46        facet_col_wrap=5,47        binary_string=binary_value,48    )49 50    return fig51 52 53def predict(img):54    try:55        img = img.reshape(1, -1)56    except:57        return "Show Your Drawing Skills"58 59    try:60        img = scaler.transform(img)61        prediction = clf.predict(img)62        return prediction[0]63    except:64        return "Train the model first"65 66 67def train_model(train_sample=5000, c=0.1, tol=0.1, solver="sage", penalty="l1"):68    X_train, X_test, y_train, y_test = train_test_split(69        X, y, train_size=train_sample, test_size=1000070    )71 72    penalty_dict = {73        "l2": ["lbfgs", "newton-cg", "newton-cholesky", "sag", "saga"],74        "l1": ["liblinear", "saga"],75        "elasticnet": ["saga"],76    }77 78    if solver not in penalty_dict[penalty]:79        return (80            "Solver not supported for the selected penalty",81            "Change the Combination",82            None,83        )84 85    global clf86    global scaler87    scaler = StandardScaler()88    X_train = scaler.fit_transform(X_train)89    X_test = scaler.transform(X_test)90 91    clf = LogisticRegression(C=c, penalty=penalty, solver=solver, tol=tol)92    clf.fit(X_train, y_train)93    sparsity = np.mean(clf.coef_ == 0) * 10094    score = clf.score(X_test, y_test)95 96    coef = clf.coef_.copy()97    scale = np.abs(coef).max()98 99    fig = px.imshow(100        np.array([coef[i].reshape(28, 28) for i in range(10)]),101        labels=dict(x="Pixel columns", y="Pixel rows"),102        title=f"Classification vector for each digit",103        range_color=[-scale, scale],104        facet_col=0,105        facet_col_wrap=5,106        facet_col_spacing=0.01,107        color_continuous_scale="RdBu",108        zmin=-scale,109        zmax=scale,110    )111 112    return score, sparsity, fig113 114 115with gr.Blocks() as demo:116    gr.Markdown("# MNIST classification using multinomial logistic + L1 ")117    gr.Markdown(118        """This interactive demo is based on the [MNIST classification using multinomial logistic + L1](https://scikit-learn.org/stable/auto_examples/linear_model/plot_sparse_logistic_regression_mnist.html#sphx-glr-auto-examples-linear-model-plot-sparse-logistic-regression-mnist-py) example from the popular [scikit-learn](https://scikit-learn.org/stable/)  library, which is a widely-used library for machine learning in Python. The primary goal of this demo is to showcase the use of logistic regression in classifying handwritten digits from the [MNIST](https://en.wikipedia.org/wiki/MNIST_database) dataset, which is a well-known benchmark dataset in computer vision. The dataset is loaded from [OpenML](https://www.openml.org/d/554), which is an open platform for machine learning research that provides easy access to a large number of datasets.119The model is trained using the scikit-learn library, which provides a range of tools for machine learning, including classification, regression, and clustering algorithms, as well as tools for data preprocessing and model evaluation. The demo calculates the score and sparsity metrics using test data, which provides insight into the model's performance and sparsity, respectively. The score metric indicates how well the model is performing, while the sparsity metric provides information about the number of non-zero coefficients in the model, which can be useful for interpreting the model and reducing its complexity.120    """121    )122 123    with gr.Tab("Explore the Data"):124        gr.Markdown("## ")125        with gr.Row():126            digit = gr.Slider(0, 9, label="Select the Digit", value=5, step=1)127            count_per_digit = gr.Slider(128                1, 10, label="Number of Images", value=10, step=1129            )130            binary_image = gr.Slider(0, 1, label="Binary Image", value=0, step=1)131 132        gen_btn = gr.Button("Show Me ")133        gen_btn.click(134            dataset_display,135            inputs=[digit, count_per_digit, binary_image],136            outputs=gr.Plot(),137        )138 139    with gr.Tab("Train Your Model"):140        gr.Markdown("# Play with the parameters to see how the model changes")141 142        gr.Markdown("## Solver and penalty")143        gr.Markdown(144            """145        Penalty | Solver146        -------|---------------147        l1 | saga 148        l2  | saga             149        """150        )151 152        with gr.Row():153            train_sample = gr.Slider(154                1000, 60000, label="Train Sample", value=5000, step=1155            )156 157            c = gr.Slider(0.1, 1, label="C", value=0.1, step=0.1)158            tol = gr.Slider(159                0.1, 1, label="Tolerance for stopping criteria.", value=0.1, step=0.1160            )161            max_iter = gr.Slider(100, 1000, label="Max Iter", value=100, step=1)162 163            penalty = gr.Dropdown(164                ["l1", "l2",], label="Penalty", value="l1"165            )166            solver = gr.Dropdown(167                ["saga"],168                label="Solver",169                value="saga",170            )171 172        train_btn = gr.Button("Train")173        train_btn.click(174            train_model,175            inputs=[train_sample, c, tol, solver, penalty],176            outputs=[177                gr.Textbox(label="Score"),178                gr.Textbox(label="Sparsity"),179                gr.Plot(),180            ],181        )182 183    with gr.Tab("Predict the Digit"):184        gr.Markdown("## Draw a digit and see the model's prediction")185        inputs = gr.Sketchpad(brush_radius=1.0)186        outputs = gr.Textbox(label="Predicted Label", lines=1)187        skecth_btn = gr.Button("Classify the Sketch")188        skecth_btn.click(predict, inputs, outputs)189 190 191demo.launch()192